Start here: what heart failure is
Heart failure is not the heart stopping. It is the heart no longer able to fill with, or eject, enough blood to meet the body's needs at a normal filling pressure. What patients notice is breathlessness — first on exertion, then lying flat, then at rest — along with swollen legs, weight gain from retained fluid, and a fatigue that no amount of rest fixes. The fluid and the fatigue are two faces of the same failure: blood backs up behind a pump that cannot move it forward, and not enough gets to the muscles in front.
The formal definition makes both halves explicit. A 2021 consensus of the American, European and Japanese heart failure societies defines it as "a clinical syndrome with symptoms and/or signs caused by a structural and/or functional cardiac abnormality and corroborated by elevated natriuretic peptide levels and/or objective evidence of pulmonary or systemic congestion" [1].
The word that matters in that sentence is syndrome. Heart failure is not a disease with a cause; it is where a lot of other diseases end up. Hypertension, coronary disease, diabetes, obesity and chronic kidney disease all funnel into it, and the same document defines a stage before symptoms — Stage A, at risk — precisely to name the people who are on that path without knowing it [1]. This review sits at the end of a cluster: hypertension, diabetes, obesity and chronic kidney disease each have their own review here, and each is upstream of this one.
The split by ejection fraction. Ejection fraction is the share of the blood in the ventricle at the end of filling that gets ejected with the next beat. The same consensus fixes the categories: HFrEF is symptomatic heart failure with an ejection fraction of 40% or less; HFmrEF — mildly reduced — is 41–49%; HFpEF — preserved — is 50% or more. There is also HFimpEF, improved: a baseline of 40% or less, a rise of at least 10 points, and a second measurement above 40% [1].
That last category exists because the disease is now sometimes reversible, which is the main news of the last decade.
Why it matters. Heart failure affects roughly 1–2% of the population worldwide [2]. It is a disease of repeated hospitalisation and shortened life, and the two halves of the ejection-fraction split have had very different fortunes: HFrEF has accumulated an extraordinary set of effective drugs, while HFpEF had essentially none until the last five years.
Why it is in the news. Two reasons, both developed below. SGLT2 inhibitors — drugs invented to lower blood sugar — turned out to reduce heart failure events across the entire ejection-fraction spectrum, including the preserved end where nothing had worked [3] [4] [5]. And GLP-1 drugs used for obesity produced the first large symptomatic benefit in the obese HFpEF phenotype [6].
Three pillars follow — measurements, medicines, and progress — with the oldest model in cardiac physiology sitting between the first two, because it explains both why the failing heart behaves as it does and why the drugs work.
Pillar 1: measurements and diagnosis
Ejection fraction, and what it does and does not tell you
Ejection fraction is measured by echocardiography: an ultrasound estimate of ventricular volume at the end of filling and at the end of ejection, with EF the fraction of the difference. It is the number that sorts patients into the categories above [1], and it is the number that determines which drugs a patient will be offered, because almost every landmark trial enrolled by it — PARADIGM-HF at 40% or less [7], DAPA-HF at 40% or less [3], EMPEROR-Preserved and DELIVER above 40% [4] [5].
Its limitation should be said plainly: EF is a ratio, not a measure of how well someone is. A patient with a small, stiff ventricle can eject 60% of a very small volume and be profoundly breathless. That is HFpEF, and it is the reason diagnosing it is harder than diagnosing HFrEF.
Natriuretic peptides
When ventricular wall stress rises, the myocardium secretes B-type natriuretic peptide. Measuring BNP or its cleavage fragment NT-proBNP therefore gives a blood test for the mechanical strain the heart is under, and it is written into the universal definition itself as corroborating evidence [1].
The peptides do three jobs. They rule out: in acute breathlessness, NT-proBNP has been pooled across 1,256 patients for diagnosis and short-term prognosis in acute destabilised heart failure [8]. They prognosticate: changes in NT-proBNP over time carry prognostic information in their own right [9]. And they support the HFpEF diagnosis, where the consensus is careful — elevated natriuretic peptides support the diagnosis, but normal levels do not exclude it [10]. Practical guidance on how to use the concentrations, and how BNP and NT-proBNP differ, has been issued separately [11] [12].
There is an elegant circularity worth noticing here, because it returns in Pillar 2: the enzyme that degrades natriuretic peptides is neprilysin, and the modern drug that improves survival in HFrEF works by inhibiting it.
Diagnosing HFpEF, which is genuinely hard
HFrEF announces itself: a dilated, poorly contracting ventricle is visible on the echo. HFpEF does not. The ventricle looks nearly normal and the problem is in filling.
The HFA-PEFF algorithm formalises the search in four steps: pre-test assessment for symptoms and the typical clinical demographics — obesity, hypertension, diabetes, older age, atrial fibrillation; then a scored echocardiographic and natriuretic-peptide step using mitral annular early diastolic velocity, E/e′ as an estimate of filling pressure, left atrial volume index, left ventricular mass index and global longitudinal strain, where ≥5 points implies definite HFpEF and ≤1 point makes it unlikely; then functional testing for intermediate scores; then a search for a specific aetiology [10]. A simpler evidence-based score exists for the same purpose [13], and Doppler and tissue-Doppler estimation of filling pressures underpins both [14] [15].
The reason the algorithm includes an exercise step is that early HFpEF can be normal at rest. Exercise haemodynamics enhance the diagnosis precisely because the abnormality only appears under load [16].
Congestion, and the functional class
Two bedside measurements carry more weight than their simplicity suggests. NYHA functional class grades symptoms from I (none with ordinary activity) to IV (symptoms at rest), and it is the variable trials enrol on: PARADIGM-HF and DAPA-HF both took class II–IV patients [7] [3], and RALES showed spironolactone improved it [17].
Congestion assessment — jugular venous pressure, oedema, lung findings — decides day-to-day management, and it has been extended by imaging: lung ultrasound now has a clinical consensus statement for use in acute and chronic heart failure [18].
Centerpiece: a simple simulatable model of the failing pump
The oldest quantitative idea in cardiac physiology is also the one that makes the rest of this review intelligible. Stretch a cardiac muscle fibre further before it contracts and it contracts harder. Scaled up to the whole ventricle: the more blood in it at the end of filling — the preload — the larger the stroke volume. That is the Frank-Starling relationship, and drawn as a curve it saturates, because there is a limit to useful stretch.
Model it as a saturating function of end-diastolic volume:
SV(EDV) = SV_max · (EDV − V₀) / (K + (EDV − V₀))
where V₀ is the filling below which nothing is ejected, SV_max is the ceiling set by contractility, and K sets how quickly the curve flattens.
What makes this the right model for heart failure is a single experimental finding. Comparing nonfailing human left ventricular myocardium against terminally failing myocardium from transplant recipients with dilated cardiomyopathy, investigators increased the preload on muscle strips and measured the force generated. In nonfailing myocardium "the force of contraction gradually increased"; in failing myocardium it "was unchanged" [19]. The failing heart is not merely operating lower on the same curve — its curve is flat. The authors traced this to the subcellular level: a failure of the myofibrils to increase their calcium sensitivity when the sarcomere is stretched [19], the length-dependent activation that in healthy muscle involves titin strain rearranging both thin and thick filaments [20] [21]. The same failure of the Frank-Starling mechanism has been documented in vivo as a cause of exercise intolerance in patients with preserved systolic function [22].
What the model explains. Four things, and each is a clinical decision.
First, why the failing heart is congested rather than simply weak. On a flat curve, extra filling buys almost no extra output. The body, sensing inadequate perfusion, activates the sympathetic and renin-angiotensin-aldosterone systems and retains salt and water — a compensation that works on a normal curve and is nearly useless on a flat one. All it achieves is higher filling pressure, which is what produces the breathlessness. This is the "vicious circle" the pathophysiology literature describes: neurohormonal activation produces volume overload and tachycardia and then further deterioration of cellular function [2].
Second, why diuretics help so much and change prognosis so little. Removing 40 mL of preload from the flat failing curve costs about 2% of stroke volume; the same move on a normal curve costs 13%. Diuresis is therefore nearly free in output terms — it relieves the congestion that causes symptoms while barely touching forward flow. But it does not change the curve. Loop diuretics are for symptoms; the four pillars are for survival [25].
Third, why the drugs are not inotropes. The intuitive fix for a weak pump is to make it squeeze harder. That is not what the four pillars do. Beta-blockers reduce contractility acutely, and they improve survival anyway [23]. What the pillars do is interrupt the neurohormonal compensation, which over months allows reverse remodelling — a smaller, better-shaped ventricle on a better curve. The figure's treated point sits at lower filling with higher output, which is the definition of a lifted curve rather than a whipped one.
Fourth, why EF can improve, and why that is a category. The 30% → 42% shift in the figure is exactly the HFimpEF definition [1]. Twenty years ago heart failure was a one-way street; the existence of a formal category for improvement is itself the news.
What the model deliberately does not do. It is a single-ventricle, single-beat relationship with no time in it: no heart rate, no afterload, no ventricular-arterial coupling, no atrial contribution, and no diastolic pressure-volume curve — which is the piece HFpEF actually turns on, since a stiff ventricle's problem is the pressure required to reach a given volume, not the volume-to-output mapping. And on the right-hand panel, the stack assumes four independent effects; the trials were run decades apart on different background therapy, and the honest treatment of that question is a formal cross-trial analysis, not multiplication [24].
Pillar 2: medicines and devices
The four pillars of HFrEF therapy
Modern HFrEF treatment is four drug classes given together, now widely called the "fantastic four" [26]. Each interrupts the neurohormonal compensation at a different point, and each was established by its own mortality trial.
Beta-blockers. Chronic sympathetic activation is toxic to the myocardium. Blocking the beta-1 adrenergic receptor reverses that, at the cost of a short-term fall in contractility that has to be titrated through. In severe heart failure with an ejection fraction under 25%, carvedilol produced a 35% decrease in the risk of death (95% CI 19 to 48%) and a 24% decrease in death or hospitalisation [23]; the benefit had been shown earlier in mild-to-moderate disease [27].
Mineralocorticoid receptor antagonists. Aldosterone drives sodium retention and myocardial fibrosis. Blocking the mineralocorticoid receptor with spironolactone in severe heart failure reduced all-cause death from 46% to 35% — a relative risk of 0.70 (95% CI 0.60–0.82) — and hospitalisation for worsening heart failure by 35% [17]. Eplerenone, a selective blocker, showed the same after myocardial infarction with left ventricular dysfunction [28].
ARNI, or an ACE inhibitor / ARB. Angiotensin II drives vasoconstriction and remodelling through the angiotensin II type-1 receptor, and blocking its production with an ACE inhibitor was the first therapy to improve survival. Sacubitril/valsartan does two things at once: it blocks that receptor and inhibits neprilysin, the enzyme that degrades natriuretic peptides — so the same peptides used in Pillar 1 as a diagnostic marker are protected and allowed to act, promoting natriuresis and vasodilation through the natriuretic peptide receptor. In PARADIGM-HF, 8,442 patients with class II–IV heart failure and an ejection fraction of 40% or less were randomised against enalapril. The trial was stopped early for overwhelming benefit: the composite of cardiovascular death or heart failure hospitalisation occurred in 21.8% versus 26.5% (hazard ratio 0.80, 95% CI 0.73–0.87), all-cause death HR 0.84 (0.76–0.93), cardiovascular death HR 0.80, and heart failure hospitalisation down 21% [7] [29].
SGLT2 inhibitors. These are the strangest of the four. SGLT2 is a glucose transporter in the kidney's proximal tubule, and the drugs were developed to lower blood sugar. In DAPA-HF, 4,744 patients with class II–IV heart failure and an ejection fraction of 40% or less received dapagliflozin or placebo on top of everything else. The primary composite occurred in 16.3% versus 21.2% (hazard ratio 0.74, 95% CI 0.65–0.85), cardiovascular death HR 0.82, all-cause death HR 0.83 — and, decisively, findings in patients with diabetes were similar to those in patients without [3]. Whatever the mechanism is, it is not glucose lowering. Symptoms and quality of life improved too [30], as did renal outcomes [31], and a meta-analysis with EMPEROR-Reduced confirmed the class effect [32].
The right-hand panel of the figure stacks these four. Taken at face value, and with the independence caveat stated there, the product is a 68% relative reduction in death from any cause — which is the argument for starting all four rather than sequencing them slowly [26] [24].
Devices
Two device therapies matter in HFrEF. Implantable cardioverter-defibrillators address the fact that a substantial share of deaths in this population are sudden and arrhythmic rather than from progressive pump failure — though the DANISH trial showed the benefit is not universal, finding no significant all-cause mortality reduction from prophylactic ICD implantation in nonischaemic systolic heart failure [33], and ICD programming itself affects outcomes [34] [35].
Cardiac resynchronisation therapy treats electrical dyssynchrony — a ventricle whose walls contract out of step — by pacing both sides. In CARE-HF, 813 patients with class III–IV heart failure and dyssynchrony on optimal drug therapy were randomised; the primary composite fell from 55% to 39% (hazard ratio 0.63, 95% CI 0.51–0.77) and death from any cause from 30% to 20% (HR 0.64, 0.48–0.85), with measured reverse remodelling: reduced end-systolic volume index and increased ejection fraction [36]. COMPANION had shown benefit with and without a defibrillator [37], and later trials extended it to milder disease [38] [39], with current indications set out in the pacing guidelines [40] [41].
HFpEF: from nothing to something
For two decades HFpEF trials failed. Spironolactone in TOPCAT was neutral overall [42]; phosphodiesterase-5 inhibition did not improve exercise capacity or clinical status [43]. The working assumption became that HFpEF was too heterogeneous to treat as one thing, and phenotype-specific treatment was proposed instead [44].
That changed with EMPEROR-Preserved: 5,988 patients with class II–IV heart failure and an ejection fraction above 40% received empagliflozin or placebo, and the composite of cardiovascular death or heart failure hospitalisation fell from 17.1% to 13.8% (hazard ratio 0.79, 95% CI 0.69–0.90), mainly through fewer hospitalisations, and again regardless of diabetes [4] [45]. DELIVER replicated it with dapagliflozin in 6,263 patients above 40% EF (HR 0.82, 0.73–0.92), with similar results above and below 60% EF [5]. Empagliflozin's benefit has been shown to run across the whole ejection-fraction spectrum [46] and across kidney function [47], and a comprehensive meta-analysis of five randomised trials now covers the class [48].
The mineralocorticoid axis came back too, in nonsteroidal form: FINEARTS-HF randomised 6,001 patients with an ejection fraction of 40% or greater to finerenone, and total worsening heart failure events plus cardiovascular death fell with a rate ratio of 0.84 (95% CI 0.74–0.95) [49].
Pillar 3: progress
Obesity-related HFpEF, and the GLP-1 drugs
A distinct obese phenotype of HFpEF has been described and characterised — different haemodynamics, different venous capacitance, different pathophysiological targets [50] [51] [52]. Until recently, describing it was all anyone could do.
STEP-HFpEF randomised 529 patients with HFpEF and a body-mass index of 30 or higher to weekly semaglutide 2.4 mg or placebo for 52 weeks, acting on the GLP-1 receptor. The dual primary endpoints both moved substantially: the Kansas City Cardiomyopathy Questionnaire clinical summary score rose 16.6 points against 8.7 (estimated difference 7.8 points, 95% CI 4.8–10.9) and body weight fell 13.3% against 2.6%. Six-minute walk distance improved by 20.3 m more, the hierarchical composite gave a win ratio of 1.72, and C-reactive protein fell 43.5% against 7.3% — and serious adverse events were lower on semaglutide, 13.3% versus 26.7% [6]. The trial was replicated in patients with type 2 diabetes [53] and pooled across the programme [54] [55] [56] [57]. Tirzepatide has since been examined for effects on circulatory overload and end-organ damage in obesity-related HFpEF [58].
Two cautions belong here. The STEP-HFpEF endpoints were symptoms, function and weight — not mortality. And the size of the CRP change raises the question of how much of the benefit is mechanical unloading versus reduced inflammation, which the programme has not settled.
Transthyretin amyloid cardiomyopathy
Some HFpEF is not HFpEF at all but an infiltrative disease that had been going undiagnosed. Wild-type transthyretin amyloidosis is now recognised as a cause of heart failure with preserved ejection fraction [59] [60], and it turns up in unexpected places — for example among elderly patients with severe aortic stenosis undergoing transcatheter valve replacement [61]. Screening approaches for everyday practice have been proposed [62].
It matters because it is treatable. Transthyretin circulates as a tetramer that must dissociate before it can misfold and deposit; tafamidis binds the tetramer and prevents that dissociation [63] [64]. In ATTR-ACT, 441 patients were randomised 2:1:2 to tafamidis 80 mg, 20 mg or placebo for 30 months: all-cause mortality was 29.5% versus 42.9% (hazard ratio 0.70, 95% CI 0.51–0.96) with a cardiovascular hospitalisation relative risk ratio of 0.68, and slower decline in both walk distance and quality of life [65] [66] [67].
Remote haemodynamic monitoring
Heart failure decompensation is preceded by rising filling pressures days before symptoms appear, which suggests measuring the pressure directly. An implanted pulmonary artery sensor allows exactly that [68]. Practice-based experience across 2,000 implanted patients [69] and a European monitoring study [70] preceded MONITOR-HF, a randomised trial of remote pulmonary artery pressure monitoring in chronic heart failure [71]. Cost-effectiveness has been analysed on both sides of the Atlantic [72] [73] [74]. Interestingly, empagliflozin's effect on pulmonary artery pressure has been measured with the same sensor [75] — one technology being used to explain another.
The cardiorenal-metabolic axis
The reason this review closes a cluster is that the organs fail together. Cardiorenal syndrome names the bidirectional failure formally [76] [77], with endothelial dysfunction proposed as the central link [78]; heart failure with both preserved and reduced ejection fraction is highly prevalent in haemodialysis patients [79]. Diabetic cardiomyopathy names the myocardial disease that diabetes causes independently of coronary artery disease [80] [81], and pre-diabetes as well as diabetes carries excess risk in HFrEF [82]. Hypertension remains the most modifiable upstream cause [83].
The therapeutic convergence is now explicit. SGLT2 inhibitors, GLP-1 receptor agonists and nonsteroidal mineralocorticoid receptor antagonists have had their combined lifetime cardiovascular, kidney and mortality benefits estimated together [84], finerenone reduces cardiovascular events in kidney disease with type 2 diabetes [85], and the ADA/EASD consensus on hyperglycaemia management now selects drugs by cardiorenal outcome rather than glucose alone [86]. Three specialties have converged on the same short list of drugs — which is the strongest evidence that they were treating one axis all along. Current practice is set out in the ESC focused update [87], the ACCF/AHA guideline [88], and an expert consensus practice update [89].
Dig deeper in lmmol
Heart failure is where the cardiometabolic-renal cluster ends, so the natural reads are the four conditions upstream of it:
- Hypertension — the most modifiable cause, and the origin of the RAAS pharmacology that became two of the four pillars.
- Diabetes — where SGLT2 inhibitors and GLP-1 agonists came from. Both arrived as glucose drugs and turned out to be heart and kidney drugs.
- Obesity — the driver behind the distinct obese HFpEF phenotype, and the reason semaglutide was tested here at all.
- Chronic kidney disease — the other half of the cardiorenal syndrome, and the review that shares three drug classes with this one.
- CKDu — the counterpoint: kidney failure in people who have none of these risk factors.
- Stroke — the other end-organ consequence of the same vascular risk cluster.
- The health reviews index collects the rest of the series.
Then move down into lmmol's graph, to the target of each pillar and each newer drug:
- Beta-1 adrenergic receptor — the beta-blocker target [23].
- Mineralocorticoid receptor — spironolactone, eplerenone and finerenone [17] [49].
- Neprilysin — the sacubitril target, and the enzyme that degrades the natriuretic peptides used to diagnose the disease [7].
- Atrial natriuretic peptide receptor 1 — where those spared peptides act.
- Sodium/glucose cotransporter 2 — the SGLT2 inhibitor target [3] [4].
- Glucagon-like peptide 1 receptor — semaglutide, in obesity-related HFpEF [6].
- Transthyretin — the tetramer tafamidis stabilises [65] [63].
- ACE and the angiotensin II type-1 receptor — the older RAAS pillar.
- Inside the sarcomere itself: SERCA2, cardiac troponin I and ryanodine receptor 2, the calcium-handling machinery whose length-dependent activation fails in the failing heart [19] [20].
- For entities without a linked static page here, use the graph index, all proteins, or all diseases rather than guessing an entity URL.